On Sun, Sep 04, 2011 at 06:14:22PM +0300, Avi Kivity wrote: > On 09/04/2011 05:54 PM, Michael S. Tsirkin wrote: > >> Way too late. And also won't work, since often the offset is > >> determined by one party and the size by another. > > > >For things like BARs, yes - but these don't need to be > >that big normally. We could add an additinal API > >that gets first/last parameters. last< first means 0 size. > >Feasible? > > Let's defer this for now. > > Does PCI actually have 64-bit addresses?
Yes. > Note that most/all cpus don't. > We may need 65-bit arithmetic for that. Jut using start/limit carefully should be enough ... > > > >> >VGA I/O addresses (including ISA aliases address - AD[15::10] are not > >> >decoded): > >> >AD[9::0] = 3B0h through 3BBh and 3C0h through 3DFh > >> > >> These "not decoded" bits mean you need to instantiate tons of > >> aliases to implement correctly. > >> I plan to add core support for that eventually. > > > >There's a flag to enable 16-bit decode actually: > >bit 4 in bridge control register. > >How does VGA work at the moment without a bridge? Ignores the ISA aliases? > >then we can do that too I think. > > > > Of course it doesn't ignore it. See the 440fx implementation, if > you disable VGA access (via the SMRAM register), vga goes away. Yes but that's for VGA RAM, right? I'm talking about the IO addresses: are tons of aliases created as you suggest? > (vga registers its legacy space as a subregion of pci_address_space(dev)) > > -- > error compiling committee.c: too many arguments to function
